MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A Beamlet-Free Algorithm for Efficient Optimization of Proton Therapy Plans Based On Monte Carlo Simulations

K Souris1*, A Barragan Montero1, G Buti1, M Cohilis1, S Wuyckens1, G Janssens2, E Sterpin1,3, J Lee1, (1) UCLouvain / MIRO, Brussels, Belgium, (2) IBA s.a., Louvain-la-Neuve, Belgium, (3) KULeuven, Leuven, Belgium

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose:
Proton therapy plan optimization is often costly in terms of computation time and memory requirements due to the necessity to compute a large number of beamlets (spot doses). This process takes time especially when Monte Carlo (MC) dose calculation is needed. Moreover, it becomes very impractical for robust optimization due to the number of beamlets growing proportionally with the number of scenarios. In this study, we designed a beamlet-free algorithm for the optimization of proton therapy plans, which reduces memory usage and computation time.


Methods:
Beamlet-based optimization algorithms typically comprise two steps: beamlet computation followed by iterative spot weight optimization until the weighted sum of all beamlets results in an optimal dose distribution.

In contrast, our beamlet-free algorithm optimizes the treatment plan with a single MC simulation. The algorithm evaluates if particles contribute either positively or negatively to the objective function during the simulation and adjusts spot weights accordingly. Thereby, the objective function is evaluated with each particle, but only considering the traversed voxels. This micro-optimization approach is similar to the stochastic gradient descent used in machine learning.

The beamlet-free algorithm was implemented in the open-source MC code MCsquare. Beamlet-based optimization was performed by the open-source optimizer MIROpt and MCsquare for beamlet calculation. Both methods were compared for the optimization of a lung proton therapy treatment.


Results:
Both methods achieved similar target coverage and organ sparing. However, the beamlet-free algorithm reduced the computation time by 59% and the RAM memory requirements by 91%.


Conclusion:
The beamlet-free algorithm reduces the computation time and memory usage. This opens the path towards online optimization for adaptive therapy but also to more comprehensive robustness solutions and more complex delivery modalities such as proton arc therapy. Moreover, the method is not limited to proton therapy and could be applied to any MC dose calculation.

Funding Support, Disclosures, and Conflict of Interest: Research collaboration with IBA s.a.

Keywords

Optimization, Protons, Monte Carlo

Taxonomy

TH- External Beam- Particle/high LET therapy: Proton therapy – dose optimization

Contact Email